A fairness-aware framework for value sharing in local electricity markets using graph and sequence learning with multi-agent reinforcement

The evolution of Local Electricity Markets (LEMs) is being driven by the integration of distributed energy resources (DERs) and peer-to-peer (P2P) energy trading. This underscores the need for practical, equitable value-allocation frameworks. Most traditional market approaches do not account for allocating revenues to prosumers and consumers sequentially, leading to suboptimal revenue distribution. This study proposes a hybrid deep learning-based framework integrating Bidirectional Long Short-Term Memory (Bi-LSTM) for energy price forecasting, Graph Neural Networks (GNN) for transaction modelling, and Multi-Agent Deep Reinforcement Learning (MADRL) for dynamic pricing optimization. A self-attention based Shapley Value computation is introduced to ensure fairness-aware revenue distribution among market participants. The results show that the framework proposed outmatches other pricing models by a significant margin. It achieves 3.3% more total energy traded and 36% better price-stability index compared to static, auction-based, and Deep Reinforcement Learning (DRL)-based baselines, while also reducing energy waste by 23% and the Gini coefficient by 32%, indicating better value equity. It achieves a 25% reduction in the Shapley-deviation metric and an increase in market participation from 72.3% to 85.6%, indicating more participation and inclusiveness. It demonstrates that the AI-driven, fairness-aware mechanisms can efficiently control market equilibrium while being transparent, stable, and low-cost. It also illustrates the profound impact AI can have in promoting local energy systems, which are sustainable, equitable, and above all, flexible.